TG-CDDPM: text-guided antimicrobial peptides generation based on conditional denoising diffusion probabilistic model.

Area of Science:

• Biotechnology
• Computational Biology
• Drug Discovery

Background:

• Antimicrobial peptides (AMPs) offer a promising alternative to antibiotics due to their broad activity, low resistance potential, and toxicity.
• Traditional AMP discovery methods are inefficient and costly.
• Existing deep generative models for AMPs often lack diversity in generated sequences.

Purpose of the Study:

• To develop a novel generative model for producing diverse and homologous antimicrobial peptides (AMPs).
• To address the limitations of existing deep learning models in generating varied AMP sequences.
• To validate the efficacy and capabilities of the proposed model in AMP discovery.

Main Methods:

• Proposed a three-stage Text-Guided Conditional Denoising Diffusion Probabilistic Model (TG-CDDPM).
• Utilized contrastive learning and inferring models in the initial stages for guided AMP generation.
• Employed a pre-trained conditional denoising diffusion probabilistic model for knowledge enrichment and fine-tuning.

Main Results:

• TG-CDDPM demonstrated competitive or superior performance compared to state-of-the-art generative models.
• The model successfully generated novel and homologous AMPs with enhanced diversity.
• Candidate AMPs identified by TG-CDDPM showed validated membrane penetration capabilities via molecular dynamics experiments.

Conclusions:

• The TG-CDDPM is an effective deep generative model for discovering diverse antimicrobial peptides.
• Text-guided generation significantly improves the quality and diversity of AMPs.
• This approach offers a powerful tool for accelerating the discovery of novel AMPs as potential antibiotics.